Introduction
Although there is unequivocal evidence for the efficacy of statins in the prevention of cardiovascular disease (CVD), there continues to be intense debate about the adverse effects of statin therapy. There are concerns both about over-prescribing and conversely that patients warranting treatment are being deterred from treatment by inaccurate reports on adverse events [
1,
2]. One potential adverse effect is an increased risk of diabetes. This increased risk was reported in Justification for the Use of Statins in Primary Prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER), a trial of rosuvastatin in people with elevated C-reactive protein, in which 3% of statin-treated and 2.4% of placebo-arm participants developed diabetes [
3]. Subsequent meta-analyses of statin trials reported increased risks [
4,
5] and observational studies also reported increased risks [
6]. In 2012 the US Food and Drug Administration required that the drug safety label was amended to indicate an increased risk [
7]. However, the aforementioned meta-analyses were not able to directly quantify the effect of statins on plasma glucose or on HbA
1c as these variables were not regularly measured in the trials examined. A recent study from Finland reported an even higher 46% increase in diabetes incidence associated with statins [
8] accompanied by both loss of insulin sensitivity and secretion. However this observational study could be subject to substantial confounding by indication.
These data have also raised concerns about potential adverse effects on diabetes in those who already have diabetes at the start of statin therapy. Recently, a combined analysis of published treatment-arm summary data on HbA
1c from statin trials concluded that there was a small average effect on HbA
1c among those with diabetes [
9]. However, this meta-analysis relied simply on published treatment-arm mean values at arbitrary points of follow-up and, as acknowledged by the authors, could not consider changes in diabetes drug use during follow-up. Thus, the estimates it used could be inaccurate and subject to competing risks bias. Furthermore, important questions remain about whether effects increase through duration of exposure. It is also unclear whether the reported effects on glycaemia reflect large effects occurring rarely in some susceptible subgroup of patients or small effects occurring commonly. Little is known about how any glycaemia effect relates to the degree of LDL-cholesterol (LDL-c) lowering or to other drug effects including liver function test changes. To address these questions individual-level data rather than meta-analyses are required.
Here we used individual-level data from a large primary prevention trial of atorvastatin 10 mg in people with existing diabetes at entry to directly test the effect of this statin on glycaemia progression as indexed by intensification of diabetes drug treatment or increasing HbA1c. We used a competing risks method that takes account of reduced CVD rates in those on statin therapy. We examined whether effects increased with duration of statin use and whether effects differed by sex. To explain possible mechanisms, we tested whether any effects were likely mediated by increased BMI and whether they related to the degree of LDL-c lowering or changes in liver function tests on treatment with statin. Finally we tested whether any adverse effect on glycaemia had an impact on the CVD risk reduction achieved with atorvastatin. Our analysis allows greater insight into the likely causality and mechanisms underlying any association and provides a practical insight allowing people with diabetes to weigh the benefits and risks of statin therapy given their condition.
Results
Of 2,838 randomised patients, 2,739 (96.5%) had at least one follow-up HbA
1c reading. The baseline characteristics of the analysis population were very similar by treatment arm, as shown in electronic supplementary material (ESM) Table
1. The mean baseline HbA
1c was slightly higher in the atorvastatin arm, by 0.05% (0.6 mmol/mol). The mean follow-up for HbA
1c change in the placebo group and atorvastatin group was 3.2 and 3.3 years, respectively; 95.3% of the participants had at least two readings and 76.0% and 47.4% had at least three or four follow-up readings, respectively. After censoring for loss to follow-up for concurrent medication or cardiovascular morbidity, the analysis population was reduced to 2,721 patients.
Time-to-event analyses of glycaemia progression
By end of follow-up, glycaemia progression occurred in 996 of 1353 placebo patients (73.6%) and 1,069 of 1368 of atorvastatin patients (78.1%) (see Table
1 for sex-specific data). Adjusting for sex, baseline age, diabetes duration and HbA
1c, and stratifying by major CVD status, the HR for the treatment effect was 1.18 (95% CI 1.08, 1.29)
p < 0.001 in both sexes combined. Glycaemia progression occurred in 78.7% of men allocated atorvastatin and 73.3% allocated placebo (HR 1.22 [95% CI 1.10, 1.35],
p < 0.001). Progression occurred in 77.0% and 74.4% of women allocated atorvastatin or placebo, respectively (HR 1.10 [95% CI 0.95, 1.29],
p = 0.197;
p = 0.289 for the diabetes by sex interaction).
Table 1
Cumulative incidence from baseline of increase in HbA1c of ≥ 0.5% (5.5 mmol/mol) or intensification of anti-glycaemic drug therapy, or both, by treatment group
Men |
1 year | 38.0 (1.6), 873 | 47.7 (1.7), 887 | 0.8 (0.3), 873 | 1.6 (0.4), 887 | 38.8 (1.7), 873 | 48.9 (1.7), 887 |
2 years | 55.7 (1.7), 844 | 61.0 (1.7), 854 | 7.1 (0.9), 844 | 6.6 (0.8), 854 | 60.1 (1.7), 844 | 64.4 (1.6), 854 |
3 years | 65.7 (1.8), 665 | 70.2 (1.8), 681 | 14.3 (1.3), 665 | 14.4 (1.3), 681 | 69.5 (1.8), 665 | 74.9 (1.7), 681 |
4 years | 69.8 (2.3), 404 | 76.0 (2.1), 417 | 31.4 (2.2), 404 | 33.9 (2.2), 417 | 78.7 (2.0), 404 | 84.2 (1.8), 417 |
Last follow-up | 61.9 (1.6), 916 | 68.3 (1.5), 929 | 34.2 (1.5), 916 | 34.5 (1.5), 929 | 73.3 (1.5), 916 | 78.7 (1.3), 929 |
Women |
1 year | 43.6 (2.5), 408 | 45.3 (2.4), 417 | 0.5 (0.3), 408 | 2.2 (0.7), 417 | 43.9 (2.5), 408 | 46.5 (2.4), 417 |
2 years | 58.5 (2.5), 402 | 61.1 (2.4), 404 | 4.7 (1.1), 402 | 6.2 (1.2), 404 | 60.0 (2.4), 402 | 63.4 (2.4), 404 |
3 years | 68.0 (2.7), 300 | 70.8 (2.6), 318 | 14.0 (2.0), 300 | 14.5 (2.0), 318 | 72.7 (2.0), 300 | 75.5 (2.4), 318 |
4 years | 73.7 (3.2), 186 | 77.1 (2.9), 205 | 26.3 (3.2), 186 | 32.2 (3.3), 205 | 80.1 (2.9), 186 | 86.3 (2.4), 205 |
Last follow-up | 66.4 (2.3), 437 | 67.0 (2.2), 439 | 28.8 (2.2), 437 | 37.1 (2.3), 439 | 74.4 (2.1), 437 | 77.0 (2.0), 439 |
These findings were robust to further adjustment by the baseline lipid levels and intensity of diabetes therapy at baseline. In the competing risk analysis the sub-hazard of glycaemia progression was similar to the main model at HR 1.21 (95% CI 1.11, 1.32), p < 0.001. In a sensitivity analysis redefining progression as intensification of diabetes therapy or at least two consecutive HbA1c readings at least 0.5% (5.5 mmol/mol) greater than baseline, a similar HR was observed (HR 1.18 [95% CI 1.07, 1.30], p = 0.001).
Overall glycaemia progression during follow-up was positively associated with younger age and lower baseline HbA
1c, lower HDL-c concentration (
p = 0.003) and higher triacylglycerol level (
p = 0.002) but was not associated with LDL-c concentration (
p = 0.720) or diabetes duration. However, there was no evidence that these or any other characteristics, including metabolic syndrome (
p = 0.243) as previously defined, heightened the effect of atorvastatin on progression [
13]. We found no evidence that the effect on glycaemia progression varied by tertile of post-treatment change in LDL-c (likelihood ratio test
p value = 0.164), creatinine kinase (
p = 0.241) or AST (
p = 0.209) but found evidence that it did vary by change in ALT. Those in the top tertile for change from baseline in ALT had an HR of 1.39 (95% CI 1.19, 1.62) for the atorvastatin effect on glycaemia progression compared with an HR of 1.14 (95% CI 1.00, 1.62) for glycaemia progression associated with atorvastatin for those in the bottom tertiles for ALT change (
p = 0.023 for the difference in treatment effect). This interaction was also present using the ALT:AST ratio.
See Table
1 and ESM Table
2 for a summary and details, respectively, of change in intensity of diabetes therapy from baseline. Overall there was no significant difference between treatment arms in the percentage of patients who progressed by at least one category of drug intensity with an HR of 1.09 (95% CI 0.96, 1.24),
p = 0.184. Thus, most of the observed effect on glycaemia progression was ascertained through an effect on HbA
1c; subsequent analyses focused on that but also adjusted for time-dependent changes in diabetes drug intensity.
Discussion
This analysis of statin effect on glycaemia within a diabetes population finds evidence for an effect of atorvastatin 10 mg daily on glycaemia progression. However the effect is modest, with a net increase in HbA1c of 0.14% (1.5 mmol/mol) by the end of follow-up. As expected there was considerable variation between patients in the net within-person change in HbA1c over time, with the effect of atorvastatin being very slight in relation to this variance in within-person change. This effect of atorvastatin on glycaemia was most apparent in the first year following initiation of therapy and, reassuringly, did not get worse through time. The effect seems to be subtle and to be a common effect rather than a large effect that occurs in a particular subgroup of patients. The only correlate of the effect was the rise in ALT during statin therapy. Importantly, we did not find any evidence that such effects on glycaemia have any material impact on the large preventive effect of atorvastatin on CVD in people with diabetes. Quantification of the glycaemic effect in diabetes is important for helping people already with diabetes, and their doctors, to arrive at an informed decision on risks and benefits of statin therapy. Of course, when making such decisions, other issues such as absolute risk of CVD, co-morbidities and other potential side effects need to be taken into consideration.
The effect on glycaemia we observed is consistent with the findings of previous meta-analyses showing increases in incident diabetes with statins [
5]. With regard to effects on measures of glycaemia there are fewer robust analyses. Small short-term intervention studies reported increases in insulin and plasma glucose or glycated albumin in hyperlipidaemic patients without diabetes [
14,
15]. More recently, Erquo et al [
9] combined treatment-arm means from published trials of those with diabetes, including unadjusted group means at 4 year follow-up, which we included as a safety analysis in the main results paper from CARDS [
10]. As acknowledged by Erquo et al, since it was not an individual-level analysis they could not assess whether effects were robust to alterations in diabetes drugs throughout the trials, nor could they assess whether or not effects increased through duration of exposure. Also, as they did not have any data on the distribution of the statin effect it remained unclear whether these average effects reflect subtle common effects or rarer extreme effects [
9].
With regard to the mechanism of this effect, the most recent meta-analysis of the effect of statins on incident diabetes was accompanied by a Mendelian randomisation analysis showing that in large association studies participants with LDL-c-lowering single-nucleotide polymorphisms in the 3-hydroxy-methylglutaryl-CoA reductase gene (with an allele effect on LDL-c of 0.06 mmol/l on average) had slightly higher rates of type 2 diabetes (OR per allele 1.02 [95% CI 1.00, 1.05]). These data were interpreted to prove a direct relationship between LDL-c lowering with statins and increased diabetes risk [
5]. In contrast we did not find any relationship with degree of LDL-c lowering in this study. Furthermore, in that analysis BMI at follow-up was increased with statin therapy in active vs placebo trials but, in contrast, we found no effect of atorvastatin on BMI despite having detailed data on this outcome. We did find some evidence that those with a greater increase in ALT following treatment were more susceptible. The METSIM (Metabolic Syndrome in Men) population-based study did report significant increases in 2 h glucose and glucose area under the receiver operating characteristic curve of an OGTT at follow-up for men without diabetes at baseline [
8]; reduced insulin sensitivity and secretion, along with a 46% increased risk of incident diabetes adjusted for baseline were also found. However, these differences could reflect confounding by indication in this observational analysis where statin recipients clearly already differed substantially from non-recipients in characteristics at baseline, before the onset of diabetes [
8].
We examined whether there are specific subgroups of people with diabetes who are most susceptible to the glycaemic effect of statins. However, we did not find any characteristics that delineated particular individuals as being susceptible to this effect. In contrast, in an analysis of atorvastatin trials in those without diabetes at initiation of therapy, those who developed new-onset type 2 diabetes were more likely to have hypertension at baseline, to be taking beta blockers and to have higher fasting glucose, BMI, white blood cell count, systolic and diastolic blood pressure, total cholesterol/ HDL-cholesterol ratio, triacylglycerol level, and lower HDL-cholesterol [
16]. In the JUPITER trial, the increased risk of diabetes with statin therapy was only observed in those with at least one diabetes risk at entry [
17]. Those with one or more major diabetes risk factors (
n = 11,508) were more likely to be women. An observational analysis from the Women’s Health Initiative reported a 1.7-fold increase in incident diabetes in postmenopausal women after starting statins [
6]. While adjustment for confounding did not reduce this risk, the potential for residual confounding to affect such observational studies remains. In contrast to these findings, we found no evidence that the effect was greater in women than in men. Sex was not associated with new-onset type 2 diabetes in an analysis of atorvastatin trials [
16].
In this study we were able to examine whether the effect of atorvastatin on glycaemia has any substantial impact on its ability to prevent CVD and we conclude that it does not. The HRs are of the same order of magnitude whether glycaemia progressed or not. Our data on the relative importance is consistent with the conclusions reached by Ridker et al [
17] in their analysis of JUPITER trial data on those without diabetes at initiation of therapy. They concluded that the cardiovascular and mortality benefits of statin therapy exceeded the diabetes hazard, including in participants at high risk of developing diabetes.
A useful aspect of our analysis that has remained unresolved in the meta-analyses so far is the role of competing risks. Previous meta-analyses were not able to assess the potential role of competing risk effects as individual-level data are needed for this. Here, such formal competing risk analyses showed that the glycaemia effect could not be attributed to this. Also the overall treatment effect is mainly driven by the change occurring in year 1. Thus, adjusting for competing risks of CVD or death has little effect on the findings.
The limitations of this study are that only atorvastatin rather than other statins and only one dose, 10 mg, was evaluated. Previous meta-analyses in individuals without diabetes have suggested that there is a dose–response effect [
5,
18]. Nonetheless, a dose of 10 mg is very commonly used in diabetes; since we have shown that many patients will achieve substantial LDL-c lowering with this dose our data have wide relevance. Additionally, the sample size precludes precise quantification of effects within subgroups. Although in the analysis of effects on HbA
1c we did adjust for time-updated changes in intensity of diabetes medications, based on the number of drugs being used, we did not adjust for dose changes within a specific category of oral drugs. However, since the estimate of effect really did not change at all, even when adjusting for time-updated change in diabetes drug intensity, it seems unlikely that it would change in a major way with the more subtle adjustment for drug dose.
Another limitation is that we did not have retinal photograph data to enable us to quantify any effects on retinopathy, which is one of the diabetes-related complications most susceptible to glycaemia.
In conclusion, although the effect of atorvastatin 10 mg on HbA1c and on glycaemia progression in patients with diabetes is statistically significant, it is very small and it does not increase with duration of statin use or have an impact on the substantial reduction in CVD risk with atorvastatin.